JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 104, NO. D20, PAGES 24,547–24,555, OCTOBER 27, 1999
A comparison of cloud top heights computed from airborne
lidar and MAS radiance data using CO2 slicing
Richard A. Frey,1 Bryan A. Baum,2 W. Paul Menzel,3 Steven A. Ackerman,1
Christopher C. Moeller,1 and James D. Spinhirne4
Abstract. Data from two instruments onboard the National Aeronautics and Space
Administration (NASA) ER-2 high-altitude aircraft have been utilized in the largest
validation study to date in assessing the accuracy of the CO2-slicing cloud height
algorithm. Infrared measurements of upwelling radiance from the MODIS (ModerateResolution Imaging Spectroradiometer) airborne simulator (MAS) were used to generate
cloud top heights and then compared to those derived from the Cloud Lidar System
(CLS), operating with dual polarization at 0.532 ␮m. The comparisons were performed for
10 flight days during the Subsonic Aircraft Contrail and Cloud Effects Special Study
(SUCCESS) field experiment during April and May 1996 which included various singlelayer and multilayer cloud conditions. Overall, the CO2-slicing method retrieved cloud
heights to within ⫾500 m and to within ⫾1500 m of the lidar heights in 32 and 64% of
the cases, respectively. From a simulation of cloud height errors as a function of various
error sources in the CO2-slicing algorithm, it was concluded that the problem of multilayer
clouds is secondary to that of proper specification of clear-sky radiances.
1.
Introduction
Clouds have a large impact on the Earth’s water and energy
budgets. Their impact on the radiation budget can result in a
heating or in a cooling of the planet, depending on the radiative properties of a cloud and its altitude [Stephens and Webster, 1981; Stephens et al., 1990]. Cloud vertical distributions,
which vary considerably [Stowe et al., 1989; Rossow et al., 1989;
Wylie and Menzel, 1989], determine diabatic heating profiles
and hence affect the general circulation of the atmosphere.
General circulation model (GCM) simulations demonstrate
the impact of changes in cloud amount and vertical structure
on atmospheric circulations [Sinha and Shine, 1995; Stubenrauch et al., 1997]. Therefore knowledge of cloud altitude and
its variation in space and time is crucial to global climate
change studies.
One important passive remote sensing method of obtaining
the altitude of middle and upper level clouds, especially transmissive clouds, is the CO2-slicing algorithm [Smith et al., 1974;
Chahine, 1974; Smith and Platt, 1978; Menzel et al., 1983].
Previous investigators have implemented the algorithm using
relatively low spatial-resolution input data such as the HighResolution Interferometer Sounder (HIS), the High-Resolution Infrared Sounder (HIRS), and the Visible-Infrared Spin
Scan Radiometer Atmospheric Sounder (VAS). Wylie and
Menzel [1989] have shown that the method retrieves cloud
heights with an overall accuracy of ⫾5 kPa, while other studies
1
Space Science and Engineering Center, University of Wisconsin,
Madison.
2
Atmospheric Sciences Division, NASA Langley Research Center,
Hampton, Virginia.
3
Satellite Applications Laboratory, NOAA NESDIS, Madison, Wisconsin.
4
NASA Goddard Space Flight Center, Greenbelt, Maryland.
Copyright 1999 by the American Geophysical Union.
Paper number 1999JD900796.
0148-0227/99/1999JD900796$09.00
have considered various associated sources of errors [Smith
and Platt, 1978; Wielicki and Coakley, 1981; Menzel et al., 1992;
Baum and Wielicki, 1994]. Other studies have compared CO2slicing cloud heights with those computed from lidar data
[Smith and Platt, 1978; Wylie and Menzel, 1989; Smith and Frey,
1990] but with a somewhat limited number of samples and over
a small geographical area.
This paper assesses the capabilities of the CO2-slicing algorithm for inferring cloud top altitude using observations from
the MODIS (Moderate-Resolution Imaging Spectroradiometer) airborne simulator (MAS) [King et al., 1996], a 50-m
resolution spectrometer flown on high-altitude aircraft. Visible
and infrared observations from the MAS are being used to
develop MODIS at-launch algorithms. The retrieved cloud top
altitudes are compared to simultaneous lidar observations
[Spinhirne and Hart, 1990]. More than 4700 collocated observations are used to validate the CO2-slicing cloud altitudes.
Investigations were conducted to explain the distribution of
cloud height differences and the most likely reason(s) for biases found between some lidar and CO2-slicing cloud heights.
2.
Data
During the Subsonic Aircraft Contrail and Cloud Effects
Special Study (SUCCESS) field campaign [Toon and MiakeLye, 1998], measurements of both clear and cloudy scenes were
taken by several instruments onboard the NASA ER-2, a highaltitude (20 km) research aircraft. Two of those instruments
were the MAS, a 50-channel spectrometer operating in the
visible, near-infrared, and infrared wavelengths, and the Cloud
Lidar System (CLS) [Spinhirne and Hart, 1990], a dualpolarization lidar operating at 0.532 ␮m. Both instruments are
designed to provide detailed information concerning cloud
characteristics such as altitude and structure. The CLS field of
view (FOV) is nominally 20 m at the surface for an aircraft
altitude of 20 km, but the raw data are usually averaged over
some period of time (between 3 and 13 s during the SUCCESS
mission) according to the quality of the lidar signal. An aver-
24,547
24,548
FREY ET AL.: CLOUD TOP HEIGHTS FROM LIDAR AND CO2 SLICING
aging time of 5 s implies a ground resolution of 20 ⫻ 1000 m
since the ER-2 travels at a speed of 200 m/s. The CLS views
only at the nadir position. The MAS instrument has a FOV of
⬃50 m at the surface and scans through ⬃90⬚ centered on
nadir (scan angle of zero degrees).
The MAS cloud mask [Ackerman et al., 1998], developed at
the Cooperative Institute for Meteorological Satellite Studies,
was used to screen the data for clear-sky FOVs.
Daily 0000 and 1200 UT temperature and moisture profiles
were obtained from National Weather Service (NWS) radiosonde data. These profiles (typically within 5 hours and 150 km
of the lidar and MAS observations) were used as input to the
CO2-slicing algorithm, but single profiles were used for multiple cloud height calculations, because of the relatively coarse
spatial and temporal resolution of NWS radiosonde launches.
A total of 15 separate profiles were used in the study.
Data were processed from 10 separate days during the
months of April and May 1996 in locations ranging from the
states of Texas to Nebraska in the central United States and
westward as far as Colorado in the Rocky Mountains. Three
other days (April 20, 27, and May 8) with collocated MAS and
CLS data were not included in the comparisons because of obvious biases between clear-sky brightness temperatures measured
by the MAS and nearby available NWS temperature profile data.
Table 1 shows the distribution of cloud height retrievals by
date and various cloud conditions. The differences between the
numbers of cloud height retrievals and the total collocations
are due to the presence of clear skies, clouds detected by the
lidar but not the MAS cloud mask and vice versa, or very small
cloud-forcing values which are smaller than the instrument
sensitivity. Note that cloud observations may fall into more
than one category and that many multilayer clouds are also
optically thin. When lidar data indicate a single-layer thick cloud,
the assumption is made that no clouds are present beneath.
3.
Technique
Cloud heights inferred from MAS and CLS data are compared in this study. The CLS lidar cloud height results are
compared to those obtained from the CO2-slicing algorithm
using MAS data. The CLS data provide cloud top and base
altitudes for a maximum of five cloud layers. The data are
provided by Spinhirne et al. [1997] and are available via the
Internet. Most details of the CO2-slicing algorithm are not
presented here, as we closely follow the method of Wylie and
Menzel [1989], excepting spectral band selection due to differ-
ent instrumentation and the use of calculated clear-sky radiances instead of observed. There are three longwave IR channel pairs found on the MAS instrument which may be used for
CO2-slicing cloud height assignment. They are the 11.0 and
13.3, 13.3 and 13.8, and the 13.8 and 14.3 ␮m pairs. Wylie and
Menzel do not use the 11.0- and 13.3-␮m channel combination.
It is used here only when a determination was made that the
cloud being viewed by the MAS was most likely an ice cloud.
The 8.5- and 11.0-␮m brightness temperature differences were
used to make this distinction because of increasing absorption
by ice across this spectral interval [Strabala et al., 1994]. Cloud
scenes with 11.0-␮m brightness temperatures at least 2K less
than the corresponding 8.5-␮m brightness temperatures are
labeled as ice. An assumption of the CO2-slicing method is that
the cloud emissivities of the channel pairs are the same [Smith
et al., 1974]. Water cloud emissivities between the 11.0- and the
13.3-␮m channels can differ to an extent that the CO2-slicing
result would be in doubt.
Since the CLS is nadir viewing, only those CO2-slicing cloud
heights from near the center of the MAS data swaths are used
in the comparison. Averaged radiance data from groups of 100
FOVs (10 ⫻ 10 boxes) centered as near as possible to nadir are
used. The average is taken to improve the signal-to-noise ratio
of the MAS radiances. Since 10 scan lines of MAS data are
gathered in about 1.5 s, there are many instances where more
than one MAS 10 ⫻ 10 FOV group falls within the averaging
period of the CLS data. In these cases the CO2-slicing cloud
heights are averaged and then compared to the lidar value.
The temperature and moisture profiles are used as input to
a forward radiative transfer model for calculation of clear-sky
radiances and for computations of spectral cloud forcing in the
CO2-slicing algorithm. Generally, the closest NWS radiosonde
data to the aircraft flight track are used; however, obvious
profile errors necessitate the use of alternative sites (but still
nearby) in a few cases. Most of the aircraft data were taken
from about 1700 to 2100 UT, so the 0000 UT profiles corresponded more closely than the 1200 UT profiles to atmospheric conditions during the flights. Time interpolation of the
profiles was not attempted. Differences between ER-2 flights
and radiosonde times would have significant effects only on
surface and near-surface temperatures in the fair weather situations in which these flights took place. The CO2-slicing algorithm is relatively immune to surface temperature errors
[Menzel et al., 1992] since upwelling radiation from the surface
contributes little to the total measured radiances of the CO2
Table 1. Distribution of Collocated CLS and MAS Data by Date and Cloud Condition
Date
Number of
Collocated
Regions
April 13
April 15
April 16
April 21
April 23
April 26
May 2
May 3
May 4
May 7
Totals
941
1101
1414
1220
809
401
901
2442
2652
1180
13061
Cloud
Retrievals
SingleLayer
Clouds
Optically
Thin
Clouds
Boundary
Layer (BL)
Clouds Only
Single-Layer,
Thin,
Non-BL
Clouds
74
5
495
1123
418
7
732
286
517
1091
4748
7
5
211
227
155
6
76
248
468
209
1612
52
5
453
642
180
7
235
138
170
318
2200
0
5
0
8
0
0
6
248
468
187
922
7
0
175
33
66
6
6
0
0
35
328
FREY ET AL.: CLOUD TOP HEIGHTS FROM LIDAR AND CO2 SLICING
24,549
Figure 1. Histogram of differences for all comparisons between Cloud Lidar System (CLS) and CO2-slicing
cloud heights in meters.
absorption channels, which are most sensitive to the middle
and upper atmosphere.
The MAS cloud mask is queried before the CO2-slicing algorithm is performed. If more than 10 FOVs are found to be cloudy,
then a cloud height is computed for that 10 ⫻ 10 FOV area.
4.
Results and Discussion
As a result of the differing times and resolutions of the CLS
and MAS cloud products, comparisons of cloud top heights
from the two data sets are somewhat problematic. The MAS
retrievals in the across-track direction, based on 10 ⫻ 10
groups of FOVs, extend ⬃250 m on either side of the nadir
position [King et al., 1996] while that of the CLS is only about
10 m [Spinhirne and Hart, 1990]. When observing broken cloud
fields, fractional cloud coverage can be very different between
the two scenes. Even when cloud fields cover the entire area of
both instruments’ FOVs, the optical depth may change from
one part of the field to another, possibly leading to substantial
differences between retrieved cloud heights. Of course, there is
also the fact that the lidar is capable of finding the very top of
the uppermost cloud layer, while the IR MAS radiances emanate mostly from some level beneath the top of the cloud,
depending on the emissive characteristics of the cloud. With
these caveats in mind, we will show that the CO2-slicing algorithm may successfully be used to characterize cloud heights
from IR data collected during the SUCCESS mission.
Figure 1 shows a histogram of all differences between cloud
top heights as retrieved from CLS data and from the CO2slicing algorithm. The difference values are calculated as the
CLS height of the top of the uppermost layer of cloud (as
reported in the CLS data) minus the collocated CO2-slicing
value. We consider the lidar values to be “truth.” The distribution of differences is approximately Gaussian with more
CO2-slicing results falling into the ⫾500 m difference category
than any other. At an altitude of 10 km in a U.S. standard
atmosphere, 500 m corresponds to ⬃2 kPa, while 2000 m is
roughly equivalent to 8 kPa. This is an encouraging result,
because this figure represents all retrievals, including multilevel cloud cases. These values match well with the error analysis performed by Wylie and Menzel [1989], where they showed
that CO2-slicing cloud heights from VAS data are accurate to
within 5 kPa.
Figure 2 is a bivariate histogram showing a comparison of
cloud top heights from both methods. Perfect agreement be-
Figure 2. Bivariate histogram showing a comparison of cloud
top heights from MODIS airborne simulator (MAS) CO2slicing and CLS lidar.
24,550
FREY ET AL.: CLOUD TOP HEIGHTS FROM LIDAR AND CO2 SLICING
Figure 3. Histogram of differences between CLS lidar and CO2-slicing cloud heights in meters. Clouds are
single-layer with tops ⬎3000 m.
tween the two would place a diagonal row of columns from top
to bottom in the diagram. The values in the two data sets agree
to within 2 km in the large majority of cases. As expected, one
sees more scatter at higher cloud altitudes because of the
higher occurrence of semitransparent cirrus clouds and multilayer clouds. Since the SUCCESS experiment focused on jet
contrails and cirrus clouds, there are few midlevel cloud cases
indicated in the figure. The “pole” seen at the bottom of the
figure represents boundary layer cloud retrievals.
With airborne lidar data, one can specify very accurately
features of the observed cloud fields [Spinhirne and Hart, 1990;
Spinhirne et al., 1982]. The geometric and relative optical thicknesses, height, and stratification of clouds may be easily and
accurately ascertained. In the following sections we will compare CO2-slicing cloud heights with those indicated by lidar as
a function of cloud type.
4.1. Single-Layer Clouds With Indications of Optical Depth
Figure 3 shows a histogram of cloud height differences between the two methods for all single-layer cloud comparisons
Figure 4. Histogram of differences between CLS lidar and CO2-slicing cloud heights in meters. Clouds are
single-layer with tops ⬎3000 m. Open bars represent optically thick clouds, while solid bars show optically thin
cloud retrievals.
FREY ET AL.: CLOUD TOP HEIGHTS FROM LIDAR AND CO2 SLICING
where cloud top heights were ⬎3000 m. Again, the shape of
the distribution is roughly Gaussian. This figure does not include boundary layer clouds, which are usually assigned a
height by simply comparing observed 11 ␮m brightness temperatures to the appropriate temperature profile. The assumption is made that these clouds radiate as blackbodies. This
procedure is necessary in the case of scenes containing only
low clouds because the difference between clear-sky and observed radiance values often falls within the instrument noise
limits, and the CO2-slicing method cannot be used. These
limits are 0.25, 0.75, 2.00, and 2.50 w/m2/steradian/␮m for the
11.0-␮m, 13.2-␮m, 14.0-␮m, and 14.2-␮m bands, respectively.
Figure 4 shows results for single-layer cloud scenes but also
distinguishes between optically thin and optically thick clouds.
When no clouds are present in the FOV of the CLS, the pulse
of light from the instrument easily reaches the ground with a
portion reflected back to the receiver on the ER-2. However, if
clouds are present and are of sufficient optical depth, the
ground signal is significantly weakened or attenuated completely. The ground signal is completely attenuated when the
total columnar aerosol and cloud optical depth exceeds 3.0
(W. D. Hart, personal communication, 1999). This information
is used as an indicator of optical thickness. Thick clouds are
defined as those for which no surface signal was present in the
CLS data. Conversely, optically thin clouds are those cases
where a cloud and a surface signal are both present in the lidar
data and the CO2-slicing method is able to retrieve a cloud
height. Since the lidar is sensitive to extremely thin clouds,
sometimes the IR cloud signal (clear - cloudy radiance) falls
within the MAS instrument noise limits and no retrieval is
possible. Notice that the class of maximum frequency is ⫾500
m and that a large majority of the results fall between ⫾1500
m for both thin and thick clouds. The shapes of the histograms
differ, however, with thick cloud errors showing a narrower
and more normal distribution. The thin cloud results also contain a slight bias toward positive differences.
These outcomes agree well with simulations. Following the
method of Baum and Wielicki [1994], the effects of several
error sources on CO2-slicing cloud heights were investigated,
including instrument noise and errors in the temperature profile used as input to radiative transfer calculations. Sets of 500
samples were constructed for 30-kPa cirrus clouds at several
effective cloud amounts, N␧ (the product of cloud fraction N
and cloud emmisivity ␧). Instrument noise was assumed to be
Gaussian with a zero mean and standard deviations equal to
the noise-equivalent radiances for the MAS channels [King et
al., 1996]. Instrument noise errors were added to the radiances
calculated from a radiative transfer model using a climatological midlatitude temperature and moisture profile. Errors in
the temperature profile were modeled as Gaussian about the
given temperature at each pressure level with a standard deviation of 2⬚.
The four histograms in Figures 5a–5d show simulations of
cloud height errors (“truth” minus realization) due to instrument noise alone and temperature profile errors and instrument noise combined for a cirrus cloud at 30 kPa and N␧ ⫽
0.3, 0.4, 0.5, and 0.6. In the thicker cloud cases (Figures 5c
and 5d), both histograms show a Gaussian distribution with
most cloud heights within 500 m of “truth.” In the case of
thinner clouds (Figures 5a and 5b), deviations due to combined
instrument noise and profile errors show a much wider distribution, primarily because the cloud signals (clear minus cloudy
radiances) are much smaller. This agrees with the results
24,551
shown in Figure 4. Given these error sources, one may expect
a distribution of retrieved cloud heights both above and below
the truth, but biases will be absent.
There were situations where errors resulted in biased cloud
heights, however. As noted above, cloud height retrievals from
several days during SUCCESS were discarded because of biases between MAS brightness temperatures and radiosonde
measurements. For example, on April 27, MAS data were
collected over northern Oklahoma, but the Norman, Oklahoma, radiosonde was unavailable. The Topeka, Kansas, profile was used instead. By comparing temperature profile data
from the CART (cloud and radiation test bed) site in Lamont,
Oklahoma, to the Topeka data, it was found that the Topeka
profile was quite a bit colder, ranging from about 10⬚ near the
surface to about 2⬚ in the upper troposphere. Figure 6 shows
the actual cloud height deviations from “truth” (lidar cloud
heights) when such a temperature bias exists, while Figure 7
shows modeled errors. Simulations are shown for the combined effects of instrument noise, a cold bias in the temperature profile of 2K (at all levels except the surface), and a
surface temperature cold bias of 5K. It is interesting to note
that at low effective cloud amounts (Figures 7a and 7b) the
errors due to instrument noise dominate errors due to biases in
the temperature profile, as seen by the spread of simulated
cloud height errors. As the effective cloud amount increases
from 0.5 (Figure 7c) to 0.6 (Figure 7d), the bias in retrieved
cloud heights due to the temperature profile errors becomes
more pronounced. However, there is still some spread in the
results due to instrument noise. In future MODIS processing,
there will be an attempt to correct for biases between temperature profiles and radiometric measurements by comparing
observed clear-sky brightness temperatures (collected during
the cloud-masking process) with calculated values within a
reasonable area and time. Frey et al. [1996] used this method
with good success in processing global HIRS 2 data. If most
biases can be removed, one may aggregate cloud height retrievals over space and time with added confidence.
4.2.
Multilayer Clouds
The CLS instrument makes it possible to validate CO2slicing cloud top heights for multilayer cloud scenes. Figure 8
shows the distribution of differences between the lidar and the
CO2-slicing results for the entire validation data set when there
are two (but only two) cloud layers separated by at least
6000 m. The ⫾500-m difference class contains the most occurrences, and the shape of the histogram is approximately Gaussian, but there is a “tail” of very low cloud heights. This tail
represents scenes where the top layer of cloud is very thin and
most of the emission measured by the MAS is coming from the
lower layer of cloud. Again, simulations help explain the shape
of the observed distribution of cloud height errors. The top two
histograms (Figures 9a and 9b) show modeled distributions of
errors due to instrument noise and the presence of low clouds
at 80 kPa, along with a 30-kPa cirrus cloud with N␧ ⫽ 0.3 and
0.6. Since climatologies suggest that about half of cloud observations are of multilayered clouds [Hahn et al., 1982, 1984], N␧
for the low clouds is set to zero for half of the realizations, but
in the other half, N␧ is allowed to vary between zero and
100%. The third and fourth histograms (Figures 9c and 9d)
show the same simulations except that random temperature
profile noise has also been included. As the upper cloud gets
optically thicker, the distribution of errors narrows and becomes approximately Gaussian with little bias, unlike the case
24,552
FREY ET AL.: CLOUD TOP HEIGHTS FROM LIDAR AND CO2 SLICING
Figure 5. Simulated cloud height errors for a cirrus cloud at 30 kPa due to instrument noise (open bars) and
atmospheric profile noise plus instrument noise (solid bars). Figures 5a–5d are for effective cloud amounts
ranging from 0.3 to 0.6.
Figure 6. Distribution of cloud height errors on April 27, 1996.
FREY ET AL.: CLOUD TOP HEIGHTS FROM LIDAR AND CO2 SLICING
Figure 7. Distribution of simulated cloud height errors when a cold bias exists in the clear-sky temperature
profile. Errors were derived using instrument noise plus a 2K cold bias in the temperature profile and a 5K
cold bias in the surface temperature. Figures 7a–7d are for effective cloud amounts of 0.3– 0.6, respectively.
Figure 8. Distribution of cloud height errors when two layers of cloud are present. The two layers are
separated by at least 6000 m.
24,553
24,554
FREY ET AL.: CLOUD TOP HEIGHTS FROM LIDAR AND CO2 SLICING
Figure 9. Distribution of simulated cloud height errors when two cloud layers are present. Top layer of
cloud is at 30 kPa, while lower layer is at 80 kPa. Figures 9a and 9b are for instrument noise and low cloud,
where the effective cloud amounts are 0.3 and 0.6, respectively. Figures 9c and 9d are the same, except
atmospheric profile noise has been added.
of temperature profile bias shown in Figure 7. Even in the case
of a very optically thin upper cloud (Figure 9a), almost 40% of
realizations fall in the ⫾500-m class. The actual cloud height
errors (Figure 8) have a distribution similar to the top, left
histogram, implying that many of the actual SUCCESS twolayer clouds came from scenes where the cirrus layer was very
optically thin. Nevertheless, the present algorithm appears to
be sufficient for accurate determination of cloud top heights in
many cases of two-layer clouds.
5.
Conclusions
More than 4700 cloud top heights measured from CLS data
and computed from MAS IR radiance data (using the CO2slicing method) during the SUCCESS field experiment were
compared. This exercise was undertaken to evaluate the CO2slicing method much as it will be implemented in the processing of the future MODIS data. In the course of the study we
have assembled the largest validation data set to date between
CO2-slicing cloud heights and lidar information, which we use
as “truth.”
Overall, the algorithm retrieved cloud heights to within
⫾500 m in 32% of cases and to within ⫾1500 m in 64% of the
cases in the 10 days of SUCCESS cloud scenes studied. This is
an encouraging result when one considers that 66% of all
comparisons came from multilayer cloud situations. One factor
to be noted about the SUCCESS data, which is both a benefit
and a liability, is that most scenes contained only cirrus and/or
underlying boundary layer clouds. This is a benefit in that
cirrus clouds pose the most difficulty when assigning cloud
height and therefore require more study, but a liability because
very few midlevel clouds are included in the validation. Even in
the cases of optically thin clouds (of which more than 72%
were multilayer), 30 and 63% of retrieved cloud top heights
were within ⫾500 and ⫾1500 m, respectively, of the lidar values.
FREY ET AL.: CLOUD TOP HEIGHTS FROM LIDAR AND CO2 SLICING
From a simulation of cloud height errors as a function of
various error sources in the CO2-slicing algorithm, we concluded that observed temperature biases between radiosonde
and radiometric data have a very negative influence on the
results. Although more comparisons are required from diverse
locations and measurement systems to increase confidence, it
appears that the problem of multilayer clouds is secondary to
that of proper specification of clear-sky radiances. Certainly,
lower clouds beneath very optically thin clouds (effective cloud
amount ⱕ0.3) lead to cloud height biases, but the overall effect
is smaller than from the use of biased clear-sky radiances. For
thicker but still transmissive clouds (effective cloud amount ⫽
0.6), the clear-sky temperature bias effect clearly dominates
that due to lower clouds. More work on the proper specification of clear-sky radiances for the purpose of cloud radiative
studies is indicated.
Acknowledgments. This research was supported by NASA under
grant NAS5-31367. The authors would also like to express gratitude to
Kathy Strabala and Liam Gumley for their helpful suggestions.
References
Ackerman, S. A., K. I. Strabala, W. P. Menzel, R. A. Frey, C. C.
Moeller, and L. E. Gumley, Discriminating clear-sky from clouds
with MODIS, J. Geophys. Res., 32,141–32,157, 1998.
Baum, B. A., and B. A. Wielicki, Cirrus cloud retrieval using infrared
sounding data: Multilevel cloud errors, J. Appl. Meteorol., 33, 107–
117, 1994.
Chahine, M. T., Remote sounding of cloudy atmospheres, 1, The single
cloud layer, J. Atmos. Sci., 31, 233–243, 1974.
Frey, R. A., S. A. Ackerman, and B. J. Soden, Climate parameters
from satellite measurements, 1, Collocated AVHRR and HIRS/2
observations of spectral greenhouse parameter, J. Clim., 9, 327–344,
1996.
Hahn, C. J., S. G. Warren, J. London, R. M. Chervin, and R. Jenne,
Atlas of simultaneous occurrence of different cloud types over the
ocean, NCAR Tech. Note, TN-201⫹STR, 212 pp., Natl. Cent. for
Atmos. Res., Boulder, Colo., 1982.
Hahn, C. J., S. G. Warren, J. London, R. M. Chervin, and R. Jenne,
Atlas of simultaneous occurrence of different cloud types over land,
NCAR Tech. Note, TN-241⫹STR, 209 pp., Natl. Cent. for Atmos.
Res., Boulder, Colo., 1984.
King, M. D., et al., Airborne scanning spectrometer for remote sensing
of cloud, aerosol, water vapor and surface properties, J. Atmos.
Oceanic Technol., 13, 777–794, 1996.
Menzel, W. P., W. L. Smith, and T. R. Stewart, Improved cloud motion
wind vector and altitude assignment using VAS, J. Appl. Meteorol.,
22, 377–384, 1983.
Menzel, W. P., D. P. Wylie, and K. I. Strabala, Seasonal and diurnal
changes in cirrus clouds as seen in four years of observations with
the VAS, J. Appl. Meteorol., 31, 370 –385, 1992.
Rossow, W. B., L. C. Gardner, and A. A. Lacis, Global seasonal cloud
24,555
variations from satellite radiance measurements, 1, Sensitivity of
analysis, J. Clim., 2, 419 – 458, 1989.
Sinha, A., and K. P. Shine, Simulated sensitivity of the Earth’s radiation budget to changes in cloud properties, Q. J. R. Meteorol. Soc.,
121, 797– 819, 1995.
Smith, W. L., and R. A. Frey, On cloud altitude determinations from
high resolution interferometer sounder (HIS) observations, J. Appl.
Meteorol., 29, 658 – 662, 1990.
Smith, W. L., and C. M. R. Platt, Comparison of satellite-deduced
cloud heights with indications from radiosonde and ground-based
laser measurements, J. Appl. Meteorol., 17, 1796 –1802, 1978.
Smith, W. L., H. M. Woolf, P. G. Abel, C. M. Hayden, M. Chalfant,
and N. Grody, Nimbus 5 sounder data processing system, 1, Measurement characteristics and data reduction procedures, NOAA
Tech. Memo., NESS 57, 99 pp., Natl. Oceanic and Atmos. Admin.,
Washington, D. C., 1974.
Spinhirne, J. D., and W. D. Hart, Cirrus structure and radiative parameters from airborne lidar and spectral radiometer observations,
Mon. Weather Rev., 118, 2329 –2343, 1990.
Spinhirne, J. D., M. Z. Hansen, and L. O. Caudill, Cloud top remote
sensing by airborne lidar, Appl. Opt., 21, 1564 –1571, 1982.
Spinhirne, J. D., W. D. Hart, and D. Hlavka, Cloud Lidar System
Onsite Quicklook Database, http://virl.gsfc.nasa.gov/⬃success/
success.html., 1997.
Stephens, G. L., and P. J. Webster, Clouds and climate: Sensitivity of
simple systems, J. Atmos. Sci., 38, 235–247, 1981.
Stephens, G. L., S. C. Tsay, J. P. W. Stackhouse, and P. Flatau, The
relevance of the microphysical and radiative properties of cirrus
clouds to climate and climatic feedback. J. Atmos. Sci., 47, 1742–
1753, 1990.
Stowe, L. L., H. Y. M. Yeh, T. F. Eck, C. G. Wellemeyer, H. L. Kyle,
and the Nimbus-7 Cloud Data Processing Team, Nimbus-7 global
cloud climatology, II, first year results, J. Clim., 2, 671–709, 1989.
Strabala, K. I., S. A. Ackerman, and W. P. Menzel, Cloud properties
inferred from 8-12-␮m data, J. Appl. Meteorol., 33, 212–229, 1994.
Stubenrauch, C. J., A. D. Del Genio, and W. B. Rossow, Implementation of subgrid cloud vertical structure inside a global circulation
model and its effect on the radiation budget, J. Clim., 10, 273–287,
1997.
Toon, O. B., and R. C. Kiake-Lye, Subsonic aircraft: Contrail and
cloud effects special study, Geophys. Res. Lett., 25, 1109 –1112, 1998.
Wielicki, B. A., and J. A. Coakley Jr., Cloud retrieval using infrared
sounder data: Error analysis, J. Appl. Meteorol., 20, 157–169, 1981.
Wylie, D. P., and W. P. Menzel, Two years of cloud cover statistics
using VAS, J. Clim. Appl. Meteorol., 2, 380 –392, 1989.
S. A. Ackerman, R. Frey, and C. C. Moeller, 1225 W. Dayton St.,
Room 219, Space Science and Engineering Center, University of Wisconsin-Madison, Madison, WI 53706. (richard.frey@ssec.wisc.edu)
B. A. Baum, Atmospheric Sciences Division, NASA Langley Research Center, Hampton, VA 23665.
W. P. Menzel, Satellite Applications Laboratory, NOAA/NESDIS,
Madison, WI 53706.
J. D. Spinhirne, NASA Goddard Space Flight Center, Greenbelt,
MD 20771.
(Received January 21, 1999; revised June 3, 1999;
accepted July 13, 1999.)
24,556